Livya powered 3D listings for dozens of property developers across North America, but every buyer question still flowed to email. The company replaced that silence with a multilingual AI concierge aware of exactly where each buyer is standing on the site.
May 8, 2026
Most property buyers research alone, at odd hours, with the volume off. They open a listing, scroll, save it, leave, return at midnight, open three more from a competitor, and never write in. By the time a buyer is willing to fill out a contact form, the developer has already lost most of the people who were curious about the same listing an hour earlier.
This is not a sales problem. It is a friction problem. You do not email. You leave.
Livya, a Quebec City property technology company, sat in the middle of this pattern. Its 3D visualization platform powers sales for dozens of developers and brokers across North America, embedded directly into each developer's own website.
Inside Livya's backend, every developer's inventory was already organized with precision: projects, buildings, floors, units, documents, surrounding district. Pricing current. Availability current. The data was there. A way to ask the data anything was not.
So inquiries went to email. Each developer ran its own website, with no shared FAQ, no standardized script, no knowledge layer stitching the portfolio together. Visitors needed answers in English, French, or Spanish, often outside business hours. Sales teams answered the ones they could. The rest faded.
Livya had built, in effect, a beautifully designed showroom with no one on the floor.
Larger players were already in conversation about adding AI to the platform; one obvious path was to co-distribute with a partner who would own the resulting system. Livya chose the other path. They would pay a builder to develop the assistant, and they would keep the intellectual property. They wanted a technology partner, not a business partner.
The decision locked in two things at once. The assistant would be theirs, and the build had to be done with the kind of care that produces something a company is willing to own. Livya brought in INTO to build it.
The instinct, when this kind of problem appears, is to bolt a chatbot onto the homepage. The instinct is wrong. A generic chatbot would have been confidently inaccurate, oblivious to what the buyer was looking at, unable to tell a one-bedroom in Quebec City from a three-bedroom in Montreal.
The first refusal: no scraping of developer websites. Developer sites change, contradict each other, and produce the noisiest possible knowledge base. Livya's own backend, where every project, building, floor, and unit already lived in clean hierarchy, would be the source of truth.
The second refusal: no single, all-purpose prompt. A buyer asking about parking is not asking the same kind of question as a buyer comparing two-bedrooms across three projects. Questions are classified first, then routed, then answered by the part of the system best equipped to answer them.
Both refusals point at the same insight. Most AI products fail not because the model is wrong, but because the system around the model has no idea what the user is looking at. Livya's platform already knew. That was the unlock.
A buyer arriving on a developer-client website is now greeted by an assistant aware of where they are: home page, project, building, floor, unit. They can ask anything, pricing, availability, amenities, neighborhood, parking, pets, financing, in English, French, or Spanish. The answers are grounded in live platform data, not in a model's general knowledge of real estate.
The assistant can also do something a generic chatbot cannot. It can search across the entire inventory. A buyer on one developer's website can ask whether any project, anywhere in the system, has a two-bedroom under four hundred thousand. Inventory that used to live behind dozens of separate websites now answers as one.
The most interesting thing about the assistant is what it refuses to do. It does not try to close.
When a buyer signals real intent, or asks something the assistant cannot confidently answer, it offers a handoff. If the buyer agrees, it asks politely for a name and either a phone number or an email. If the buyer declines, it moves on without pushing. If the buyer has already shared their details, the assistant remembers and does not ask again. Each handoff produces a structured email to the developer's sales team with the full conversation attached.
The fastest way to make an AI feel cheap is to have it ask for an email address every third sentence. The assistant is not a replacement for the salesperson. It is a way of making sure the salesperson is talking to a buyer who has already moved through the questions only a system can patiently answer.
Livya wanted to operate the assistant themselves. What that produced is closer to an operating console than a settings panel. A product manager can edit how the assistant handles financing questions, run the change in a live environment against real test cases, watch the result, save it with full version history, and trace any production answer back to its reasoning in a single click.
This sounds like an engineering detail. It is a strategic one. The companies that compound advantage from AI are the companies that own the dial.
The first developer went live on July 1, 2025. Routine questions, pricing, availability, financing, neighborhood, now resolve inside the assistant, in the buyer's preferred language, at any hour. The questions that genuinely require a salesperson reach one, with the full context of the conversation already in the email body. Inventory that had been spread across as many websites as developers became searchable through a single conversational surface.
The technology is the least interesting part of the story. What buyers wanted was not artificial intelligence. They wanted to ask one question without writing an email about it. They wanted to know whether dogs were allowed before they kept scrolling. They wanted to compare two projects without opening five tabs. The system finally answers the question the buyer was already asking silently. Once that loop is closed, the rest is just better versions of itself.
With expertise in strategy and product management, Sebastien helps organizations integrate AI in their business operations and services.
Most AI assistants fail because the system around the model has no idea what the user is looking at. Livya's 3D property platform already knew, in precise hierarchy: project, building, floor, unit. That structure became the substrate for a multilingual buyer concierge that turned dozens of fragmented developer websites into one searchable surface, and email-only inquiries into qualified conversations.
Thirty minutes. We'll tell you what to build, what not to build, and what it would take.